Audio signal encoding and decoding method using learning model, training method of learning model, and encoder and decoder that perform the methods

A training method of a learning model used to decode an audio signal, comprising:

• extracting a first residual signal and a first linear prediction coefficient by decoding a bitstream received from an encoder;
• generating an audio signal from the first residual signal using the first linear prediction coefficient;
• generating second linear prediction coefficients and a second residual signal from the audio signal;
• comparing the first linear prediction coefficient with a third linear prediction coefficient obtained by inputting the second linear prediction coefficient into the learning model;
• and
• updating parameters of the learning model based on the comparison result.

An audio signal encoding method comprising:

• extracting an original residual signal and an original linear prediction coefficient from an audio signal;
• encoding the original residual signal;
• and
• converting the encoded original residual signal and the original linear prediction coefficient into a bitstream,
• wherein the original linear prediction coefficient is used as target data for training a learning model used by a decoder, and
• the learning model generates an output linear prediction coefficient from an input linear prediction coefficient, and is trained using the original linear prediction coefficient as target data.